📄 swap_sse2.s

📁 Optimized GotoBLAS libraries
💻 S
字号:
/*********************************************************************//*                                                                   *//*             Optimized BLAS libraries                              *//*                     By Kazushige Goto <kgoto@tacc.utexas.edu>     *//*                                                                   *//* Copyright (c) The University of Texas, 2005. All rights reserved. *//* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING  *//* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF      *//* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE,              *//* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY  *//* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF     *//* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO   *//* THE USE OF THE SOFTWARE OR DOCUMENTATION.                         *//* Under no circumstances shall University be liable for incidental, *//* special, indirect, direct or consequential damages or loss of     *//* profits, interruption of business, or related expenses which may  *//* arise from use of Software or Documentation, including but not    *//* limited to those resulting from defects in Software and/or        *//* Documentation, or loss or inaccuracy of data of any kind.         *//*********************************************************************/#define ASSEMBLER#include "common.h"#ifndef WINDOWS_ABI#define M	ARG1#define X	ARG4#define INCX	ARG5#define Y	ARG6#define INCY	ARG2#else#define M	ARG1#define X	ARG2#define INCX	ARG3#define Y	ARG4#define INCY	%rbx#endif#define XX	%r10#define YY	%r11#ifdef BARCELONA#define	PREFETCHW	prefetchw#define	PREFETCHSIZE	(8 * 16)#endif	PROLOGUE	PROFCODE#ifndef WINDOWS_ABI	movq	8(%rsp), INCY#else	pushq	%rbx	movq	48(%rsp), X	movq	56(%rsp), INCX	movq	64(%rsp), Y	movq	72(%rsp), INCY#endif	SAVEREGISTERS	leaq	(, INCX, SIZE), INCX	leaq	(, INCY, SIZE), INCY	cmpq	$SIZE, INCX	jne	.L40	cmpq	$SIZE, INCY	jne	.L40	testq	$SIZE, Y	je	.L10	movsd	0 * SIZE(X), %xmm0	movsd	0 * SIZE(Y), %xmm8	movsd	%xmm8, 0 * SIZE(X)	movsd	%xmm0, 0 * SIZE(Y)	addq	$1 * SIZE, X	addq	$1 * SIZE, Y	decq	M	jle	.L19	ALIGN_4.L10:	testq	$SIZE, X	jne	.L20	movq	M,  %rax	sarq	$4, %rax	jle	.L13	movapd	 0 * SIZE(X), %xmm0	movapd	 2 * SIZE(X), %xmm1	movapd	 4 * SIZE(X), %xmm2	movapd	 6 * SIZE(X), %xmm3	movapd	 0 * SIZE(Y), %xmm8	movapd	 2 * SIZE(Y), %xmm9	movapd	 4 * SIZE(Y), %xmm10	movapd	 6 * SIZE(Y), %xmm11	movapd	 8 * SIZE(X), %xmm4	movapd	10 * SIZE(X), %xmm5	movapd	12 * SIZE(X), %xmm6	movapd	14 * SIZE(X), %xmm7	movapd	 8 * SIZE(Y), %xmm12	movapd	10 * SIZE(Y), %xmm13	movapd	12 * SIZE(Y), %xmm14	movapd	14 * SIZE(Y), %xmm15	decq	%rax	jle .L12	ALIGN_3.L11:#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 0)(X)#endif	movapd	%xmm8, 	 0 * SIZE(X)	movapd	%xmm9, 	 2 * SIZE(X)	movapd	%xmm10,	 4 * SIZE(X)	movapd	%xmm11,	 6 * SIZE(X)	movapd	16 * SIZE(Y), %xmm8	movapd	18 * SIZE(Y), %xmm9	movapd	20 * SIZE(Y), %xmm10	movapd	22 * SIZE(Y), %xmm11#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 0)(Y)#endif	movapd	%xmm0, 	 0 * SIZE(Y)	movapd	%xmm1, 	 2 * SIZE(Y)	movapd	%xmm2, 	 4 * SIZE(Y)	movapd	%xmm3, 	 6 * SIZE(Y)	movapd	16 * SIZE(X), %xmm0	movapd	18 * SIZE(X), %xmm1	movapd	20 * SIZE(X), %xmm2	movapd	22 * SIZE(X), %xmm3#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 8)(X)#endif	movapd	%xmm12,	 8 * SIZE(X)	movapd	%xmm13,	10 * SIZE(X)	movapd	%xmm14,	12 * SIZE(X)	movapd	%xmm15,	14 * SIZE(X)	movapd	24 * SIZE(Y), %xmm12	movapd	26 * SIZE(Y), %xmm13	movapd	28 * SIZE(Y), %xmm14	movapd	30 * SIZE(Y), %xmm15#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 8)(Y)#endif	movapd	%xmm4, 	 8 * SIZE(Y)	movapd	%xmm5, 	10 * SIZE(Y)	movapd	%xmm6, 	12 * SIZE(Y)	movapd	%xmm7, 	14 * SIZE(Y)	movapd	24 * SIZE(X), %xmm4	movapd	26 * SIZE(X), %xmm5	movapd	28 * SIZE(X), %xmm6	movapd	30 * SIZE(X), %xmm7	subq	$-16 * SIZE, Y	subq	$-16 * SIZE, X	decq	%rax	jg	.L11	ALIGN_3.L12:	movapd	%xmm8, 	 0 * SIZE(X)	movapd	%xmm9, 	 2 * SIZE(X)	movapd	%xmm10,	 4 * SIZE(X)	movapd	%xmm11,	 6 * SIZE(X)	movapd	%xmm0, 	 0 * SIZE(Y)	movapd	%xmm1, 	 2 * SIZE(Y)	movapd	%xmm2, 	 4 * SIZE(Y)	movapd	%xmm3, 	 6 * SIZE(Y)	movapd	%xmm12,	 8 * SIZE(X)	movapd	%xmm13,	10 * SIZE(X)	movapd	%xmm14,	12 * SIZE(X)	movapd	%xmm15,	14 * SIZE(X)	movapd	%xmm4, 	 8 * SIZE(Y)	movapd	%xmm5, 	10 * SIZE(Y)	movapd	%xmm6, 	12 * SIZE(Y)	movapd	%xmm7, 	14 * SIZE(Y)	addq	$16 * SIZE, X	addq	$16 * SIZE, Y	ALIGN_3.L13:	movq	M,  %rax	andq	$8, %rax	jle	.L14	ALIGN_3	movapd	0 * SIZE(X), %xmm0	movapd	2 * SIZE(X), %xmm1	movapd	4 * SIZE(X), %xmm2	movapd	6 * SIZE(X), %xmm3	movapd	0 * SIZE(Y), %xmm8	movapd	2 * SIZE(Y), %xmm9	movapd	4 * SIZE(Y), %xmm10	movapd	6 * SIZE(Y), %xmm11	movapd	%xmm8, 	0 * SIZE(X)	movapd	%xmm9, 	2 * SIZE(X)	movapd	%xmm10,	4 * SIZE(X)	movapd	%xmm11,	6 * SIZE(X)	movapd	%xmm0, 	0 * SIZE(Y)	movapd	%xmm1, 	2 * SIZE(Y)	movapd	%xmm2, 	4 * SIZE(Y)	movapd	%xmm3, 	6 * SIZE(Y)	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L14:	movq	M,  %rax	andq	$4, %rax	jle	.L15	ALIGN_3	movapd	0 * SIZE(X), %xmm0	movapd	2 * SIZE(X), %xmm1	movapd	0 * SIZE(Y), %xmm8	movapd	2 * SIZE(Y), %xmm9	movapd	%xmm8, 	0 * SIZE(X)	movapd	%xmm9, 	2 * SIZE(X)	movapd	%xmm0, 	0 * SIZE(Y)	movapd	%xmm1, 	2 * SIZE(Y)	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L15:	movq	M,  %rax	andq	$2, %rax	jle	.L16	ALIGN_3	movapd	0 * SIZE(X), %xmm0	movapd	0 * SIZE(Y), %xmm8	movapd	%xmm8, 	0 * SIZE(X)	movapd	%xmm0, 	0 * SIZE(Y)	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L16:	movq	M,  %rax	andq	$1, %rax	jle	.L19	ALIGN_3	movsd	0 * SIZE(X), %xmm0	movsd	0 * SIZE(Y), %xmm8	movsd	%xmm8, 	0 * SIZE(X)	movsd	%xmm0, 	0 * SIZE(Y)	ALIGN_3.L19:	xorq	%rax,%rax	RESTOREREGISTERS#ifdef WINDOWS_ABI	popq	%rbx#endif	ret	ALIGN_3.L20:	movapd	-1 * SIZE(X), %xmm0	movq	M,  %rax	sarq	$4, %rax	jle	.L23	movapd	  1 * SIZE(X), %xmm1	movapd	  3 * SIZE(X), %xmm2	movapd	  5 * SIZE(X), %xmm3	movapd	  7 * SIZE(X), %xmm4	movapd	  9 * SIZE(X), %xmm5	movapd	 11 * SIZE(X), %xmm6	movapd	 13 * SIZE(X), %xmm7	movapd	  0 * SIZE(Y), %xmm8	movapd	  2 * SIZE(Y), %xmm9	movapd	  4 * SIZE(Y), %xmm10	movapd	  6 * SIZE(Y), %xmm11	movapd	  8 * SIZE(Y), %xmm12	movapd	 10 * SIZE(Y), %xmm13	movapd	 12 * SIZE(Y), %xmm14	movapd	 14 * SIZE(Y), %xmm15	movlpd	%xmm8,   0 * SIZE(X)	decq	%rax	jle .L22	ALIGN_4.L21:#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 0)(X)#endif	SHUFPD_1 %xmm1, %xmm0	SHUFPD_1 %xmm2, %xmm1	SHUFPD_1 %xmm3, %xmm2	SHUFPD_1 %xmm4, %xmm3	movapd	%xmm0,   0 * SIZE(Y)	movapd	%xmm1,   2 * SIZE(Y)	movapd	%xmm2,   4 * SIZE(Y)	movapd	%xmm3,   6 * SIZE(Y)	movapd	 15 * SIZE(X), %xmm0	movapd	 17 * SIZE(X), %xmm1	movapd	 19 * SIZE(X), %xmm2	movapd	 21 * SIZE(X), %xmm3#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 0)(Y)#endif	SHUFPD_1 %xmm5, %xmm4	SHUFPD_1 %xmm6, %xmm5	SHUFPD_1 %xmm7, %xmm6	SHUFPD_1 %xmm0, %xmm7	SHUFPD_1 %xmm9,   %xmm8	SHUFPD_1 %xmm10,  %xmm9	SHUFPD_1 %xmm11,  %xmm10	SHUFPD_1 %xmm12,  %xmm11	movapd	%xmm8,   1 * SIZE(X)	movapd	%xmm9,   3 * SIZE(X)	movapd	%xmm10,  5 * SIZE(X)	movapd	%xmm11,  7 * SIZE(X)	movapd	 16 * SIZE(Y), %xmm8	movapd	 18 * SIZE(Y), %xmm9	movapd	 20 * SIZE(Y), %xmm10	movapd	 22 * SIZE(Y), %xmm11#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 8)(X)#endif	SHUFPD_1 %xmm13,  %xmm12	SHUFPD_1 %xmm14,  %xmm13	SHUFPD_1 %xmm15,  %xmm14	SHUFPD_1 %xmm8,   %xmm15	movapd	%xmm12,  9 * SIZE(X)	movapd	%xmm13, 11 * SIZE(X)	movapd	%xmm14, 13 * SIZE(X)	movapd	%xmm15, 15 * SIZE(X)#ifdef BARCELONA	PREFETCHW	(PREFETCHSIZE + 8)(Y)#endif	movapd	 24 * SIZE(Y), %xmm12	movapd	 26 * SIZE(Y), %xmm13	movapd	 28 * SIZE(Y), %xmm14	movapd	 30 * SIZE(Y), %xmm15	movapd	%xmm4,   8 * SIZE(Y)	movapd	%xmm5,  10 * SIZE(Y)	movapd	%xmm6,  12 * SIZE(Y)	movapd	%xmm7,  14 * SIZE(Y)	movapd	 23 * SIZE(X), %xmm4	movapd	 25 * SIZE(X), %xmm5	movapd	 27 * SIZE(X), %xmm6	movapd	 29 * SIZE(X), %xmm7	subq	$-16 * SIZE, X	subq	$-16 * SIZE, Y	decq	%rax	jg	.L21	ALIGN_3.L22:	SHUFPD_1 %xmm1, %xmm0	SHUFPD_1 %xmm2, %xmm1	SHUFPD_1 %xmm3, %xmm2	SHUFPD_1 %xmm4, %xmm3	movapd	%xmm0,   0 * SIZE(Y)	movapd	%xmm1,   2 * SIZE(Y)	movapd	%xmm2,   4 * SIZE(Y)	movapd	%xmm3,   6 * SIZE(Y)	movapd	 15 * SIZE(X), %xmm0	SHUFPD_1 %xmm5, %xmm4	SHUFPD_1 %xmm6, %xmm5	SHUFPD_1 %xmm7, %xmm6	SHUFPD_1 %xmm0, %xmm7	SHUFPD_1 %xmm9,   %xmm8	SHUFPD_1 %xmm10,  %xmm9	SHUFPD_1 %xmm11,  %xmm10	SHUFPD_1 %xmm12,  %xmm11	SHUFPD_1 %xmm13,  %xmm12	SHUFPD_1 %xmm14,  %xmm13	SHUFPD_1 %xmm15,  %xmm14	movapd	%xmm4,   8 * SIZE(Y)	movapd	%xmm5,  10 * SIZE(Y)	movapd	%xmm6,  12 * SIZE(Y)	movapd	%xmm7,  14 * SIZE(Y)	movapd	%xmm8,   1 * SIZE(X)	movapd	%xmm9,   3 * SIZE(X)	movapd	%xmm10,  5 * SIZE(X)	movapd	%xmm11,  7 * SIZE(X)	movapd	%xmm12,  9 * SIZE(X)	movapd	%xmm13, 11 * SIZE(X)	movapd	%xmm14, 13 * SIZE(X)	movhpd	%xmm15, 15 * SIZE(X)	addq	$16 * SIZE, X	addq	$16 * SIZE, Y	ALIGN_3.L23:	movq	M,  %rax	andq	$8, %rax	jle	.L24	ALIGN_3	movapd	1 * SIZE(X), %xmm1	movapd	3 * SIZE(X), %xmm2	movapd	5 * SIZE(X), %xmm3	movapd	7 * SIZE(X), %xmm4	movapd	0 * SIZE(Y), %xmm8	movapd	2 * SIZE(Y), %xmm9	movapd	4 * SIZE(Y), %xmm10	movapd	6 * SIZE(Y), %xmm11	SHUFPD_1 %xmm1, %xmm0	SHUFPD_1 %xmm2, %xmm1	SHUFPD_1 %xmm3, %xmm2	SHUFPD_1 %xmm4, %xmm3	movlpd	%xmm8,  0 * SIZE(X)	SHUFPD_1 %xmm9,  %xmm8	SHUFPD_1 %xmm10, %xmm9	SHUFPD_1 %xmm11, %xmm10	movapd	%xmm8,  1 * SIZE(X)	movapd	%xmm9,  3 * SIZE(X)	movapd	%xmm10, 5 * SIZE(X)	movhpd	%xmm11, 7 * SIZE(X)	movapd	%xmm0,  0 * SIZE(Y)	movapd	%xmm1,  2 * SIZE(Y)	movapd	%xmm2,  4 * SIZE(Y)	movapd	%xmm3,  6 * SIZE(Y)	movapd	%xmm4, %xmm0	addq	$8 * SIZE, X	addq	$8 * SIZE, Y	ALIGN_3.L24:	movq	M,  %rax	andq	$4, %rax	jle	.L25	ALIGN_3	movapd	1 * SIZE(X), %xmm1	movapd	3 * SIZE(X), %xmm2	movapd	0 * SIZE(Y), %xmm8	movapd	2 * SIZE(Y), %xmm9	SHUFPD_1 %xmm1, %xmm0	SHUFPD_1 %xmm2, %xmm1	movlpd	%xmm8, 0 * SIZE(X)	SHUFPD_1 %xmm9, %xmm8	movapd	%xmm8, 1 * SIZE(X)	movhpd	%xmm9, 3 * SIZE(X)	movapd	%xmm0, 0 * SIZE(Y)	movapd	%xmm1, 2 * SIZE(Y)	movapd	%xmm2, %xmm0	addq	$4 * SIZE, X	addq	$4 * SIZE, Y	ALIGN_3.L25:	movq	M,  %rax	andq	$2, %rax	jle	.L26	ALIGN_3	movapd	1 * SIZE(X), %xmm1	SHUFPD_1 %xmm1, %xmm0	movapd	0 * SIZE(Y), %xmm8	movlpd	%xmm8, 0 * SIZE(X)	movhpd	%xmm8, 1 * SIZE(X)	movapd	%xmm0, 0 * SIZE(Y)	addq	$2 * SIZE, X	addq	$2 * SIZE, Y	ALIGN_3.L26:	movq	M,  %rax	andq	$1, %rax	jle	.L29	ALIGN_3	movsd	0 * SIZE(X), %xmm0	movsd	0 * SIZE(Y), %xmm8	movsd	%xmm8, 	0 * SIZE(X)	movsd	%xmm0, 	0 * SIZE(Y)	ALIGN_3.L29:	xorq	%rax,%rax	RESTOREREGISTERS#ifdef WINDOWS_ABI	popq	%rbx#endif	ret	ALIGN_3.L40:	movq	X, XX	movq	Y, YY	movq	M,  %rax	sarq	$3, %rax	jle	.L45	ALIGN_3.L41:	movsd	0 * SIZE(X), %xmm0	addq	INCX, X	movsd	0 * SIZE(Y), %xmm8	addq	INCY, Y	movsd	0 * SIZE(X), %xmm1	addq	INCX, X	movsd	0 * SIZE(Y), %xmm9	addq	INCY, Y	movsd	0 * SIZE(X), %xmm2	addq	INCX, X	movsd	0 * SIZE(Y), %xmm10	addq	INCY, Y	movsd	0 * SIZE(X), %xmm3	addq	INCX, X	movsd	0 * SIZE(Y), %xmm11	addq	INCY, Y	movsd	0 * SIZE(X), %xmm4	addq	INCX, X	movsd	0 * SIZE(Y), %xmm12	addq	INCY, Y	movsd	0 * SIZE(X), %xmm5	addq	INCX, X	movsd	0 * SIZE(Y), %xmm13	addq	INCY, Y	movsd	0 * SIZE(X), %xmm6	addq	INCX, X	movsd	0 * SIZE(Y), %xmm14	addq	INCY, Y	movsd	0 * SIZE(X), %xmm7	addq	INCX, X	movsd	0 * SIZE(Y), %xmm15	addq	INCY, Y	movsd	%xmm8, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm0, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm9, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm1, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm10, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm2, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm11, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm3, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm12, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm4, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm13, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm5, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm14, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm6, 0 * SIZE(YY)	addq	INCY, YY	movsd	%xmm15, 0 * SIZE(XX)	addq	INCX, XX	movsd	%xmm7, 0 * SIZE(YY)	addq	INCY, YY	decq	%rax	jg	.L41	ALIGN_3.L45:	movq	M,  %rax	andq	$7, %rax	jle	.L47	ALIGN_3.L46:	movsd	(X), %xmm0	movsd	(Y), %xmm8	movsd	%xmm8, (X)	movsd	%xmm0, (Y)	addq	INCX, X	addq	INCY, Y	decq	%rax	jg	.L46	ALIGN_3.L47:	xorq	%rax, %rax	RESTOREREGISTERS#ifdef WINDOWS_ABI	popq	%rbx#endif	ret	EPILOGUE
⌨️ 快捷键说明

复制代码 Ctrl + C
搜索代码 Ctrl + F
全屏模式 F11
切换主题 Ctrl + Shift + D
显示快捷键 ?
增大字号 Ctrl + =
减小字号 Ctrl + -